Show HN: Advanced Code Search with Tree-sitter AST and Qdrant Vector DB

5 points by anonentity_lc a day ago

For the past couple of months, I’ve been building a tool that enables natural language search over large codebases using Tree-Sitter for syntax parsing and Qdrant for vector-based retrieval. https://app.repogram.com

### How It Works

- Tree-Sitter is used to parse syntax trees and extract high-quality vector embeddings of code. - These embeddings are stored in Qdrant, enabling fast similarity search across your entire repo. - A combination of re-ranking processes refine search results, producing highly relevant answers to code-related questions.

The results have been incredibly accurate, thanks to the quality and structured nature of the embeddings. I've been working through the TypeORM issue backlog with some excellent results.

Repogram doesn’t just help you search code, it enables a shared, evolving knowledge base:

- Questions are converted into public, protected, or private threads that remain searchable by repo members. - Over time, this builds up tribal knowledge that acts as an interactive alternative to traditional documentation. - Threads can be optionally summarized into searchable pages, making it easier to navigate discussions.

### Getting Started

- Enter a GitHub repo URL or name. - Sync the repo to generate embeddings. - Query your codebase using natural language.

### Currently supported languages:

- TypeScript - Javascript - Rust - Python - Go - Java - PHP - Swift - Ruby

...with additional support for Markdown. I’ll be adding support for more languages over the coming weeks.

To get started, log in with GitHub to start exploring. I’ve already synced a few large open-source libraries like TypeORM, Fastify, and BullMQ, so feel free to experiment—or sync your own projects with a fine-grained PAT via the UI.

It's still very much a work in progress but the UI is fully functional. If you hit any bugs, let me know and I’ll patch them as quickly as possible. Would love to hear your thoughts! Feedback, ideas, and feature suggestions are all welcome.